FastHunyuan Open-Source Video Generation Model - Rapidly Produce High-Quality Videos in 6 Steps with Speed Up by Approximately 8 Times

Fasthunyuan

Developed by FastVideo

FastHunyuan is the accelerated version of HunyuanVideo, requiring only 6 diffusion sampling steps to generate high-quality videos, achieving approximately an 8x speed improvement compared to the original version.

Text-to-Video

Safetensors

Open Source License:Other #Text-to-Video Generation #Efficient Inference #Low-Step Sampling

Downloads 94

Release Time : 12/16/2024

Model Overview

FastHunyuan is a text-to-video generation model that achieves efficient video generation capabilities through consistency distillation training.

Model Features

Efficient Video Generation

Only 6 diffusion sampling steps are needed to generate high-quality videos, achieving approximately an 8x speed improvement compared to the original HunyuanVideo's 50-step sampling.

High-Quality Output

While reducing the sampling steps, it still maintains high-quality video generation.

Low VRAM Requirements

Can run inference on a single RTX4090 GPU, supporting NF4 and LLM-INT8 quantized inference via BitsAndBytes.

Model Capabilities

Text-to-Video Generation

High-Quality Video Synthesis

Efficient Inference

Use Cases

Video Creation

Short Video Generation

Quickly generate short video content based on text descriptions.

Generates videos with 720x1280 resolution and 125 frames.

Creative Content Production

Used for rapid production of creative video content such as advertisements and social media.

🚀 FastHunyuan Model Card

FastHunyuan is an accelerated video generation model that significantly speeds up the video sampling process while maintaining high - quality output. It offers an efficient solution for text - to - video tasks.

🚀 Quick Start

Clone the Fastvideo repository and follow the inference instructions in the README.
Alternatively, you can perform inference on FastHunyuan using the official Hunyuan Video repository by setting the shift to 17, steps to 6, resolution to 720X1280X125, and cfg bigger than 6. We find that a large CFG scale generally leads to faster videos.

✨ Features

FastHunyuan is an accelerated HunyuanVideo model. It can sample high - quality videos with 6 diffusion steps, which brings around 8X speed up compared to the original HunyuanVideo with 50 steps.

📦 Installation

No specific installation steps are provided in the original README.

📚 Documentation

Model Details

Developed by: Hao AI Lab
License: tencent - hunyuan - community
Distilled from: HunyuanVideo
Github Repository: https://github.com/hao-ai-lab/FastVideo

Training details

FastHunyuan is consistency distillated on the MixKit dataset with the following hyperparameters:

Batch size: 16
Resolution: 720x1280
Num of frames: 125
Train steps: 320
GPUs: 32
LR: 1e - 6
Loss: huber

Evaluation

We provide some qualitative comparison between FastHunyuan 6 - step inference v.s. the original Hunyuan with 6 - step inference:

FastHunyuan 6 step	Hunyuan 6 step

Memory requirements

Please check our GitHub repo for details. https://github.com/hao-ai-lab/FastVideo

For inference, we can perform inference on FastHunyuan using a single RTX4090. We now support NF4 and LLM - INT8 quantized inference using BitsAndBytes for FastHunyuan. With NF4 quantization, inference can be performed on a single RTX 4090 GPU, requiring just 20GB of VRAM.

For Lora Finetune, the minimum hardware requirements are:

40 GB GPU memory each for 2 GPUs with lora
30 GB GPU memory each for 2 GPUs with CPU offload and lora.

📄 License

The model is released under the tencent - hunyuan - community license. You can find more details in the LICENSE file.

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご